search

r-spatial / dtwSat

Time-Weighted Dynamic Time Warping for satellite image time series analysis
https://www.victor-maus.com/dtwSat/
GNU General Public License v3.0
130 stars 39 forks source link

Different geographical locations for Temporal Patterns and Actual Raster Bands #48

Closed Randy6668 closed 3 years ago

Randy6668 commented 4 years ago

Dear all, I'm currently attempting to do a TWDTW classification for a region without any known samples (Region A), but I do have sample points from a nearby region (and associated time series data; Region B). I'm attempting to build the temporal patterns data from Region B and use them to classify Region A. But from what I understand, twdtwApply requires both training and validation rasters to be the same.

Is there any workabout for this? Been looking through other available functions, but could not find much so far. Thanks!

TheSl0thzilla commented 4 years ago

I run into the same problem. Have you found a solution?

vwmaus commented 3 years ago

Build the patterns for region B, where you have samples, then classify region A using these patterns. That is no problem for TWDTW. Differently from ML approaches, TWDTW generalises well across space and time, as long as the regions are still within the same clime zone. If possible please report your experience here.